In the article http://nlp.fast.ai/classification/2018/05/15/introducting-ulmfit.html it notes that ULMFiT “better solves problems like…Finding documents relevant to a legal case”.
I have an opportunity to work on such a problem and would love to learn if there is a tangible use case to back this statement or if this is more of a theoretical statement. If there is an experiment that was done for this particular use case, could it be shared?
Many thanks in advance.
@jeremy, can you help with this question? I’m trying to work with Microsoft on this within Azure and it would be great to have a legal use case to back me up. Have you anything outside of the IMDB use case that we can use?
There’s a mention of it in the course, or you could search the forum to find it.
(Having said that, it’s a fairly direct ULMFiT classification exercise.)
I searched the forum using the following search criteria, “ULMFiT legal”, and didn’t find anything. There were some multi-lingual posts but nothing pertaining to multi-label classification on legal documents.
I’ll keep looking…
You might be looking for these parts in Jeremy’s lesson 4, as transcribed by @PoonamV :
watch from min: 03:01
📝 Deep Learning Lesson 4 Notes - look for “Classifying legal text”
Thank you @jolackner. I plan on watching that video today…I hope they go over multi-label classification using ULMFiT. I’m also looking for a dataset of legal documents to practice with and will post back if I find anything relevant.
Well, Im collecting all the references mentioned in the videos, currently Im on this one… so
I found this https://www.comp.nus.edu.sg/~kanmy/courses/6101_1810/ dont know if the capture is from item 13 in the list of Student Projects. Here another thread which links to img there 📝 Deep Learning Lesson 4 Notes
But it seems that was only a jpg, thought here is this https://arxiv.org/abs/1904.06470 which seems to be the final paper.